[moconnor]

Actually this shows why OpenAI matters. Google have been training and refining Transformer architectures for years; how unlikely is it nobody tried training a language model at this scale or larger with similar results?

Yet from Google we heard nothing. Which is the optimal decision for them - they only lose by blowing the whistle.

[sgt101]

A lot of people have results similar to this - but most people generating a paragraph of slightly_weird_but_plausible_if_you_read_quickly text using a primped version of BERT one time out of 25 regarded it as more or less pointless. But journalists don't.

This would be ok if this is the first time that anyone had a media go wild over AI story. But actually this has happened 10000 times this year already.

[skybrian]

Seems like the way it worked is that the blog post was discussed here and on Twitter and many people thought it was interesting. Then some journalists picked it up and wrote about it.

That much is nothing out of the ordinary. It is interesting (at least to those of us who aren't natural language researchers) so why shouldn't we talk about it? Why shouldn't journalists write about it?

Inevitably their mildly controversial decision to hold some data back got a lot of people discussing whether it was necessary. Which is also perfectly okay.

So, in the end, the complaint is just about why people don't have smarter takes on things. I don't know what to tell you; that's just how social media works sometimes.

[sgt101]

I'd shrug and move on, but the problem is that I believe that these flaps about AI are distracting attention from the real concerns and forces that are having a serious impact on people now.

The distortion of public debate caused by community exclusiveness on social platforms, by the curation and manipulation of social feeds and by the dynamics of online debate where the loudest and angriest voices dominate is one place that we could do with some focus.

Another place is the management of simple models - plain Jane stuff like a learned classifier - people are making these with Python and R and releasing them into infrastructures and apps and we don't know what they are and where they are and how they are interacting.

Instead we have wizard of oz style stories to distract us from who's actually hiding behind the curtain. If we fall for this then we may find ourselves living in a vicious totalitarian society with no obvious way out of it.

Journalists should write about it in an informed and professional way, that's fine. But they need to write about stories that are impactful and important, and if they were to write about this one in this way ("text scrambler makes a pretty good paragraph one out of 30 tries, has no idea of what is going on") they would get no clicks (there will now be a second wave of follow-ups like that to ride on the coattails of the story). Instead they have to make it sound like robots are going to take children from schools and experiment on them live on TV, and this makes them famous and rich.

There is no real revision of the story because the follow on stories disappear from view while search engines and other journalists use the original hysteria. Look at what happened with the two negotiating bots at facebook (the game was to negotiate to get books and balls, the bots tended to use a short hand to negotiate rather than the english they were trained on) This was "Facebook researchers have to pull the plug on AI that they no longer understand", and that is the narrative that we will have on that story more or less forever.